Boards and CEOs hiring a Chief Data & AI Officer want the same thing: proof, quickly, that the function will pay for itself. The first 90 days decide whether that proof ever arrives. Here is how I would spend them.
Days 1–30: outcomes and mandate
Start with the business, not the technology. Sit down with the CEO and the CFO and agree on the three business outcomes the function exists to move, each with a number attached and a date by which it should move. Then agree on what comes with those outcomes: the decision rights, the budget, and the authority to stop work that does not serve them.
Agree, in the same conversation, on how value will be measured and reported. If the finance team does not recognize the number, it is not a result.
Then take inventory. List every AI initiative in flight, who owns it, what it costs, and whether it has a path to production. Most organizations discover that they have plenty of experiments and very little AI in production. That inventory becomes the honest starting point for everything that follows.
Days 31–60: foundations, first bets, and risk
Stand up, or reset, the Data & AI governance council, with business owners in the chair rather than technology leaders. Governance built around business outcomes is what lets an enterprise scale AI without scaling risk; governance built around control slows everything down.
Assess data readiness for the priority use cases only. A full enterprise data assessment takes a year and answers questions nobody asked. Readiness for the three outcomes takes weeks and tells you what to fix first.
Put AI risk on the council’s agenda from its first meeting, and treat it as a board conversation rather than a compliance checkbox. That means an inventory of every model and agent in use, a named owner and defined permissions for each, audit trails, and a short report the audit committee can read. Where the industry is regulated, build the operating model on a recognized framework, NIST’s AI Risk Management Framework or ISO 42001, so the board’s questions have answers before they are asked. I have stood this up for clients in insurance, financial services, and life sciences; the framework matters less than the habit of reporting.
Then pick the first bets: two or three use cases with a clear path to production and a number attached. Fewer, bigger, and finished beats many and abandoned.
Days 61–90: operating model and people
Publish the operating model. Say what is central and what is federated, how agents are governed, where humans decide and where agents act, and where the hand-offs and controls sit. Be explicit about the boundary with the CIO and CTO: who owns the platforms, who owns the data products, who owns the models, and how the three of you make decisions together. Fund process redesign and change management alongside the models, because an agent dropped into an old process automates the old process.
Lay out the talent plan: the roles to hire, the people to develop, and the communities of practice that will carry the work after the program ends.
Finally, report the first results against the outcome charter agreed in the first month. Not a status deck. Results, in the CFO’s numbers, with what comes next.
What the board should expect at day 90
A signed outcome charter. A governance council that has met and decided something, with AI risk on its agenda. A funded portfolio of two or three use cases with production dates. A published operating model with a clear boundary to the CIO. A talent plan. A first number that moved. And a list of what was stopped, because a leader who has not stopped anything in 90 days has not yet used the authority the role came with.